My criterion for valid benchmarks.

Recently I have fielded many inquiries related to building the fastest servers available.

So, in reply. I am publishing the list of criterion I adhere to. These are requirements for the server systems that I build and sell for production web or file servers. My standards of quality and moral values in this regard will not waiver based on the lure of easy wealth or time constraints. If I do not have the time to focus the requisite attention on building a server right, it will not be built. I will not cut corners on quality. I build each server by hand. I build each system in Long Beach, California. All sales are direct from Evanscomputers.com. If you find a system for sale with my name on it elsewhere, it’s either used (not likely), or an imitation.

1. Stock and reasonably available production parts. Using prototype parts is out of the question. Engineering samples are legitimate if you can later reproduce the result with a production part.
2. Standard clock speeds. Overclocking may be a wonderful option for a gaming system. When you plan to use the system as a production server, that has to be online and operational 24×7 365 days of the year ( or as close to that as possible ) overclocking is not an option. Not to mention that it voids your warranties for the processor and motherboard in most cases.
3. A reproducible production grade build of the system. If you spend 25 hours modifying the system and can not reproduce that with every server you manufacture for sale to your clients, it does not apply.
4. An off the shelf operating system. If you modify the operating system for the benchmark results, and you can not or will not do so for your production systems, it would not apply. If it is a tuning or selection of components that you can and do reproduce for a production system, that’s perfectly acceptable. See rule 2.
5. Reproducible results of the benchmark. If you perform the benchmark once and can not reproduce the results using the same criterion again, using the same production grade system, it is not a valid benchmark.
6. Test tools that are readily available and produce consistent results. I personally prefer Phoronix Test Suite. The tests are standardized and produce consistent results for the same versions of the software. There are other test suites available that provide consistent reliable benchmark results.
7. A system that is configured correctly and securely for a production server environment. SELinux or equivalent enabled in enforcing targeted mode at the very least.

All that said, when following the above guidelines, I build the fastest servers that exist. Faster than Dell 2, 4, 8 or even 16 core servers. Faster than Fujitsu-Siemens Dual Xeons. Faster than even HP Proliant servers. I do not build the least expensive servers available, only the best.

For a listing of the results, the best, the worst, and the mediocre, of my benchmarking results visit:
Phoronix Evanscomputers.com Results

I am never the fastest for every benchmark. I am usually the fastest for Super-Pi and Apache-Build benchmarks. I do get bested on occasion. Currently I am the fastest for Super-Pi to 1 Million digits using the criterion stated above with results that are consistently reproducible and published at Phoronix. Dealer inquiries are not welcome here. Comments and your business are! : )

Leave a Reply

Your email address will not be published. Required fields are marked *